4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 struct kmem_cache *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
97 lru_page_max = pages / 2;
99 lru_page_max = (pages / 4) * 3;
102 /* initialize lru data */
103 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
104 sbi->ll_cache.ccc_lru_max = lru_page_max;
105 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
106 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
107 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
109 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
110 SBI_DEFAULT_READAHEAD_MAX);
111 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
112 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
113 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
114 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
115 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
117 ll_generate_random_uuid(uuid);
118 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
119 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
121 spin_lock(&ll_sb_lock);
122 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
123 spin_unlock(&ll_sb_lock);
125 sbi->ll_flags |= LL_SBI_VERBOSE;
126 #ifdef ENABLE_CHECKSUM
127 sbi->ll_flags |= LL_SBI_CHECKSUM;
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
137 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
141 /* metadata statahead is enabled by default */
142 sbi->ll_sa_max = LL_SA_RPC_DEF;
143 cfs_atomic_set(&sbi->ll_sa_total, 0);
144 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
145 cfs_atomic_set(&sbi->ll_agl_total, 0);
146 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 spin_lock(&ll_sb_lock);
158 cfs_list_del(&sbi->ll_list);
159 spin_unlock(&ll_sb_lock);
160 OBD_FREE(sbi, sizeof(*sbi));
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = 0;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_device *obd;
171 struct obd_capa *oc = NULL;
172 struct obd_statfs *osfs = NULL;
173 struct ptlrpc_request *request = NULL;
174 struct obd_connect_data *data = NULL;
175 struct obd_uuid *uuid;
176 struct md_op_data *op_data;
177 struct lustre_md lmd;
179 int size, err, checksum;
182 obd = class_name2obd(md);
184 CERROR("MD %s: not setup or attached\n", md);
198 /* indicate the features supported by this client */
199 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
200 OBD_CONNECT_ATTRFID |
201 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
202 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
203 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
204 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
205 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
206 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
207 OBD_CONNECT_EINPROGRESS |
208 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
209 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
210 OBD_CONNECT_MAX_EASIZE |
211 OBD_CONNECT_FLOCK_DEAD;
213 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
214 data->ocd_connect_flags |= OBD_CONNECT_SOM;
216 #ifdef HAVE_LRU_RESIZE_SUPPORT
217 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
218 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
220 #ifdef CONFIG_FS_POSIX_ACL
221 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
224 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
225 /* flag mdc connection as lightweight, only used for test
226 * purpose, use with care */
227 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
229 data->ocd_ibits_known = MDS_INODELOCK_FULL;
230 data->ocd_version = LUSTRE_VERSION_CODE;
232 if (sb->s_flags & MS_RDONLY)
233 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
234 if (sbi->ll_flags & LL_SBI_USER_XATTR)
235 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
237 #ifdef HAVE_MS_FLOCK_LOCK
238 /* force vfs to use lustre handler for flock() calls - bug 10743 */
239 sb->s_flags |= MS_FLOCK_LOCK;
241 #ifdef MS_HAS_NEW_AOPS
242 sb->s_flags |= MS_HAS_NEW_AOPS;
245 if (sbi->ll_flags & LL_SBI_FLOCK)
246 sbi->ll_fop = &ll_file_operations_flock;
247 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
248 sbi->ll_fop = &ll_file_operations;
250 sbi->ll_fop = &ll_file_operations_noflock;
253 data->ocd_connect_flags |= OBD_CONNECT_REAL;
254 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
255 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
257 data->ocd_brw_size = MD_MAX_BRW_SIZE;
259 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
261 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
262 "recovery, of which this client is not a "
263 "part. Please wait for recovery to complete,"
264 " abort, or time out.\n", md);
267 CERROR("cannot connect to %s: rc = %d\n", md, err);
271 sbi->ll_md_exp->exp_connect_data = *data;
273 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
274 LUSTRE_SEQ_METADATA);
276 CERROR("%s: Can't init metadata layer FID infrastructure, "
277 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
281 /* For mount, we only need fs info from MDT0, and also in DNE, it
282 * can make sure the client can be mounted as long as MDT0 is
284 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
285 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
286 OBD_STATFS_FOR_MDT0);
288 GOTO(out_md_fid, err);
290 /* This needs to be after statfs to ensure connect has finished.
291 * Note that "data" does NOT contain the valid connect reply.
292 * If connecting to a 1.8 server there will be no LMV device, so
293 * we can access the MDC export directly and exp_connect_flags will
294 * be non-zero, but if accessing an upgraded 2.1 server it will
295 * have the correct flags filled in.
296 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
297 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
298 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
299 valid != CLIENT_CONNECT_MDT_REQD) {
302 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
303 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
304 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
305 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
306 "feature(s) needed for correct operation "
307 "of this client (%s). Please upgrade "
308 "server or downgrade client.\n",
309 sbi->ll_md_exp->exp_obd->obd_name, buf);
310 OBD_FREE(buf, PAGE_CACHE_SIZE);
311 GOTO(out_md_fid, err = -EPROTO);
314 size = sizeof(*data);
315 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
316 KEY_CONN_DATA, &size, data, NULL);
318 CERROR("%s: Get connect data failed: rc = %d\n",
319 sbi->ll_md_exp->exp_obd->obd_name, err);
320 GOTO(out_md_fid, err);
323 LASSERT(osfs->os_bsize);
324 sb->s_blocksize = osfs->os_bsize;
325 sb->s_blocksize_bits = log2(osfs->os_bsize);
326 sb->s_magic = LL_SUPER_MAGIC;
327 sb->s_maxbytes = MAX_LFS_FILESIZE;
328 sbi->ll_namelen = osfs->os_namelen;
329 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
331 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
332 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
333 LCONSOLE_INFO("Disabling user_xattr feature because "
334 "it is not supported on the server\n");
335 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
338 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
340 sb->s_flags |= MS_POSIXACL;
342 sbi->ll_flags |= LL_SBI_ACL;
344 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
346 sb->s_flags &= ~MS_POSIXACL;
348 sbi->ll_flags &= ~LL_SBI_ACL;
351 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
352 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
353 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
354 LCONSOLE_INFO("client is set as remote by default.\n");
357 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
358 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
359 LCONSOLE_INFO("client claims to be remote, but server "
360 "rejected, forced to be local.\n");
364 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
365 LCONSOLE_INFO("client enabled MDS capability!\n");
366 sbi->ll_flags |= LL_SBI_MDS_CAPA;
369 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
370 LCONSOLE_INFO("client enabled OSS capability!\n");
371 sbi->ll_flags |= LL_SBI_OSS_CAPA;
374 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
375 sbi->ll_flags |= LL_SBI_64BIT_HASH;
377 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
378 sbi->ll_md_brw_size = data->ocd_brw_size;
380 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
382 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
383 LCONSOLE_INFO("Layout lock feature supported.\n");
384 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
387 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
388 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
389 LCONSOLE_INFO("%s: disabling xattr cache due to "
390 "unknown maximum xattr size.\n", dt);
392 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
393 sbi->ll_xattr_cache_enabled = 1;
397 obd = class_name2obd(dt);
399 CERROR("DT %s: not setup or attached\n", dt);
400 GOTO(out_md_fid, err = -ENODEV);
403 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
404 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
405 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
406 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
407 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
408 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
409 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
410 OBD_CONNECT_MAXBYTES |
411 OBD_CONNECT_EINPROGRESS |
412 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
413 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
415 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
416 data->ocd_connect_flags |= OBD_CONNECT_SOM;
418 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
419 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
420 * disabled by default, because it can still be enabled on the
421 * fly via /proc. As a consequence, we still need to come to an
422 * agreement on the supported algorithms at connect time */
423 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
425 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
426 data->ocd_cksum_types = OBD_CKSUM_ADLER;
428 data->ocd_cksum_types = cksum_types_supported_client();
431 #ifdef HAVE_LRU_RESIZE_SUPPORT
432 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
434 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
435 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
437 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
438 "ocd_grant: %d\n", data->ocd_connect_flags,
439 data->ocd_version, data->ocd_grant);
441 obd->obd_upcall.onu_owner = &sbi->ll_lco;
442 obd->obd_upcall.onu_upcall = cl_ocd_update;
444 data->ocd_brw_size = DT_MAX_BRW_SIZE;
446 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
449 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
450 "recovery, of which this client is not a "
451 "part. Please wait for recovery to "
452 "complete, abort, or time out.\n", dt);
455 CERROR("%s: Cannot connect to %s: rc = %d\n",
456 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
460 sbi->ll_dt_exp->exp_connect_data = *data;
462 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
463 LUSTRE_SEQ_METADATA);
465 CERROR("%s: Can't init data layer FID infrastructure, "
466 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
470 mutex_lock(&sbi->ll_lco.lco_lock);
471 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
472 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
473 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
474 mutex_unlock(&sbi->ll_lco.lco_lock);
476 fid_zero(&sbi->ll_root_fid);
477 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
479 CERROR("cannot mds_connect: rc = %d\n", err);
480 GOTO(out_lock_cn_cb, err);
482 if (!fid_is_sane(&sbi->ll_root_fid)) {
483 CERROR("%s: Invalid root fid "DFID" during mount\n",
484 sbi->ll_md_exp->exp_obd->obd_name,
485 PFID(&sbi->ll_root_fid));
486 GOTO(out_lock_cn_cb, err = -EINVAL);
488 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
490 sb->s_op = &lustre_super_operations;
491 #if THREAD_SIZE >= 8192 /*b=17630*/
492 sb->s_export_op = &lustre_export_operations;
496 * XXX: move this to after cbd setup? */
497 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
498 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
499 valid |= OBD_MD_FLRMTPERM;
500 else if (sbi->ll_flags & LL_SBI_ACL)
501 valid |= OBD_MD_FLACL;
503 OBD_ALLOC_PTR(op_data);
505 GOTO(out_lock_cn_cb, err = -ENOMEM);
507 op_data->op_fid1 = sbi->ll_root_fid;
508 op_data->op_mode = 0;
509 op_data->op_capa1 = oc;
510 op_data->op_valid = valid;
512 err = md_getattr(sbi->ll_md_exp, op_data, &request);
515 OBD_FREE_PTR(op_data);
517 CERROR("%s: md_getattr failed for root: rc = %d\n",
518 sbi->ll_md_exp->exp_obd->obd_name, err);
519 GOTO(out_lock_cn_cb, err);
522 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
523 sbi->ll_md_exp, &lmd);
525 CERROR("failed to understand root inode md: rc = %d\n", err);
526 ptlrpc_req_finished(request);
527 GOTO(out_lock_cn_cb, err);
530 LASSERT(fid_is_sane(&sbi->ll_root_fid));
531 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
532 sbi->ll_flags & LL_SBI_32BIT_API),
534 md_free_lustre_md(sbi->ll_md_exp, &lmd);
535 ptlrpc_req_finished(request);
537 if (root == NULL || IS_ERR(root)) {
539 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
540 #ifdef CONFIG_FS_POSIX_ACL
542 posix_acl_release(lmd.posix_acl);
543 lmd.posix_acl = NULL;
546 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
548 CERROR("lustre_lite: bad iget4 for root\n");
552 err = ll_close_thread_start(&sbi->ll_lcq);
554 CERROR("cannot start close thread: rc %d\n", err);
558 #ifdef CONFIG_FS_POSIX_ACL
559 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
560 rct_init(&sbi->ll_rct);
561 et_init(&sbi->ll_et);
565 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
566 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
567 KEY_CHECKSUM, sizeof(checksum), &checksum,
571 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
572 KEY_CACHE_SET, sizeof(sbi->ll_cache),
573 &sbi->ll_cache, NULL);
575 sb->s_root = d_make_root(root);
576 if (sb->s_root == NULL) {
577 CERROR("%s: can't make root dentry\n",
578 ll_get_fsname(sb, NULL, 0));
579 GOTO(out_root, err = -ENOMEM);
581 #ifdef HAVE_DCACHE_LOCK
582 sb->s_root->d_op = &ll_d_ops;
585 sbi->ll_sdev_orig = sb->s_dev;
587 /* We set sb->s_dev equal on all lustre clients in order to support
588 * NFS export clustering. NFSD requires that the FSID be the same
590 /* s_dev is also used in lt_compare() to compare two fs, but that is
591 * only a node-local comparison. */
592 uuid = obd_get_uuid(sbi->ll_md_exp);
594 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
600 if (proc_lustre_fs_root != NULL) {
601 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
604 CERROR("%s: could not register mount in lprocfs: "
605 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
615 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
617 obd_disconnect(sbi->ll_dt_exp);
618 sbi->ll_dt_exp = NULL;
619 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
620 obd_zombie_barrier();
622 obd_fid_fini(sbi->ll_md_exp->exp_obd);
624 obd_disconnect(sbi->ll_md_exp);
625 sbi->ll_md_exp = NULL;
634 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
638 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
640 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
641 KEY_MAX_EASIZE, &size, lmmsize, NULL);
643 CERROR("Get max mdsize error rc %d \n", rc);
648 void ll_dump_inode(struct inode *inode)
650 struct ll_d_hlist_node *tmp;
651 int dentry_count = 0;
653 LASSERT(inode != NULL);
655 ll_d_hlist_for_each(tmp, &inode->i_dentry)
658 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
659 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
660 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
663 void lustre_dump_dentry(struct dentry *dentry, int recur)
665 struct list_head *tmp;
668 LASSERT(dentry != NULL);
670 list_for_each(tmp, &dentry->d_subdirs)
673 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
674 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
675 dentry->d_name.len, dentry->d_name.name,
676 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
677 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
678 dentry->d_flags, dentry->d_fsdata, subdirs);
679 if (dentry->d_inode != NULL)
680 ll_dump_inode(dentry->d_inode);
685 list_for_each(tmp, &dentry->d_subdirs) {
686 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
687 lustre_dump_dentry(d, recur - 1);
691 void client_common_put_super(struct super_block *sb)
693 struct ll_sb_info *sbi = ll_s2sbi(sb);
696 #ifdef CONFIG_FS_POSIX_ACL
697 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
698 et_fini(&sbi->ll_et);
699 rct_fini(&sbi->ll_rct);
703 ll_close_thread_shutdown(sbi->ll_lcq);
707 cfs_list_del(&sbi->ll_conn_chain);
709 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
710 obd_disconnect(sbi->ll_dt_exp);
711 sbi->ll_dt_exp = NULL;
712 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
714 obd_zombie_barrier();
716 lprocfs_unregister_mountpoint(sbi);
718 obd_fid_fini(sbi->ll_md_exp->exp_obd);
719 obd_disconnect(sbi->ll_md_exp);
720 sbi->ll_md_exp = NULL;
725 void ll_kill_super(struct super_block *sb)
727 struct ll_sb_info *sbi;
732 if (!(sb->s_flags & MS_ACTIVE))
736 /* we need restore s_dev from changed for clustred NFS before put_super
737 * because new kernels have cached s_dev and change sb->s_dev in
738 * put_super not affected real removing devices */
740 sb->s_dev = sbi->ll_sdev_orig;
741 sbi->ll_umounting = 1;
746 char *ll_read_opt(const char *opt, char *data)
752 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
753 if (strncmp(opt, data, strlen(opt)))
755 if ((value = strchr(data, '=')) == NULL)
759 OBD_ALLOC(retval, strlen(value) + 1);
761 CERROR("out of memory!\n");
765 memcpy(retval, value, strlen(value)+1);
766 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
770 static inline int ll_set_opt(const char *opt, char *data, int fl)
772 if (strncmp(opt, data, strlen(opt)) != 0)
778 /* non-client-specific mount options are parsed in lmd_parse */
779 static int ll_options(char *options, int *flags)
782 char *s1 = options, *s2;
788 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
791 CDEBUG(D_SUPER, "next opt=%s\n", s1);
792 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
797 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
802 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
807 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
812 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
817 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
822 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
823 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
825 /* Ignore deprecated mount option. The client will
826 * always try to mount with ACL support, whether this
827 * is used depends on whether server supports it. */
828 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
829 "mount option 'acl'.\n");
832 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
834 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
835 "mount option 'noacl'.\n");
839 #warning "{no}acl options have been deprecated since 1.8, please remove them"
841 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
846 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
851 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
857 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
862 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
867 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
872 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
877 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
882 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
887 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
892 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
897 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
902 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
907 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
913 s2 = strchr(s1, ',');
921 void ll_lli_init(struct ll_inode_info *lli)
923 lli->lli_inode_magic = LLI_INODE_MAGIC;
925 lli->lli_ioepoch = 0;
926 lli->lli_maxbytes = MAX_LFS_FILESIZE;
927 spin_lock_init(&lli->lli_lock);
928 lli->lli_posix_acl = NULL;
929 lli->lli_remote_perms = NULL;
930 mutex_init(&lli->lli_rmtperm_mutex);
931 /* Do not set lli_fid, it has been initialized already. */
932 fid_zero(&lli->lli_pfid);
933 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
934 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
935 cfs_atomic_set(&lli->lli_open_count, 0);
936 lli->lli_mds_capa = NULL;
937 lli->lli_rmtperm_time = 0;
938 lli->lli_pending_och = NULL;
939 lli->lli_mds_read_och = NULL;
940 lli->lli_mds_write_och = NULL;
941 lli->lli_mds_exec_och = NULL;
942 lli->lli_open_fd_read_count = 0;
943 lli->lli_open_fd_write_count = 0;
944 lli->lli_open_fd_exec_count = 0;
945 mutex_init(&lli->lli_och_mutex);
946 spin_lock_init(&lli->lli_agl_lock);
947 lli->lli_has_smd = false;
948 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
949 lli->lli_clob = NULL;
951 init_rwsem(&lli->lli_xattrs_list_rwsem);
952 mutex_init(&lli->lli_xattrs_enq_lock);
954 LASSERT(lli->lli_vfs_inode.i_mode != 0);
955 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
956 mutex_init(&lli->lli_readdir_mutex);
957 lli->lli_opendir_key = NULL;
959 lli->lli_def_acl = NULL;
960 spin_lock_init(&lli->lli_sa_lock);
961 lli->lli_opendir_pid = 0;
963 sema_init(&lli->lli_size_sem, 1);
964 lli->lli_size_sem_owner = NULL;
965 lli->lli_symlink_name = NULL;
966 init_rwsem(&lli->lli_trunc_sem);
967 mutex_init(&lli->lli_write_mutex);
968 init_rwsem(&lli->lli_glimpse_sem);
969 lli->lli_glimpse_time = 0;
970 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
971 lli->lli_agl_index = 0;
972 lli->lli_async_rc = 0;
973 lli->lli_volatile = false;
975 mutex_init(&lli->lli_layout_mutex);
978 static inline int ll_bdi_register(struct backing_dev_info *bdi)
980 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
982 bdi->name = "lustre";
983 return bdi_register(bdi, NULL, "lustre-%d",
984 atomic_inc_return(&ll_bdi_num));
987 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
989 struct lustre_profile *lprof = NULL;
990 struct lustre_sb_info *lsi = s2lsi(sb);
991 struct ll_sb_info *sbi;
992 char *dt = NULL, *md = NULL;
993 char *profilenm = get_profile_name(sb);
994 struct config_llog_instance *cfg;
995 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
996 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1000 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1008 /* client additional sb info */
1009 lsi->lsi_llsbi = sbi = ll_init_sbi();
1011 cfs_module_put(THIS_MODULE);
1016 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1018 GOTO(out_free, err);
1020 err = bdi_init(&lsi->lsi_bdi);
1022 GOTO(out_free, err);
1023 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1024 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1025 err = ll_bdi_register(&lsi->lsi_bdi);
1027 GOTO(out_free, err);
1029 sb->s_bdi = &lsi->lsi_bdi;
1030 #ifndef HAVE_DCACHE_LOCK
1031 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1032 sb->s_d_op = &ll_d_ops;
1035 /* Generate a string unique to this super, in case some joker tries
1036 to mount the same fs at two mount points.
1037 Use the address of the super itself.*/
1038 cfg->cfg_instance = sb;
1039 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1040 cfg->cfg_callback = class_config_llog_handler;
1041 /* set up client obds */
1042 err = lustre_process_log(sb, profilenm, cfg);
1044 CERROR("Unable to process log: %d\n", err);
1045 GOTO(out_free, err);
1048 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1049 lprof = class_get_profile(profilenm);
1050 if (lprof == NULL) {
1051 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1052 " read from the MGS. Does that filesystem "
1053 "exist?\n", profilenm);
1054 GOTO(out_free, err = -EINVAL);
1056 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1057 lprof->lp_md, lprof->lp_dt);
1059 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1061 GOTO(out_free, err = -ENOMEM);
1062 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1064 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1066 GOTO(out_free, err = -ENOMEM);
1067 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1069 /* connections, registrations, sb setup */
1070 err = client_common_fill_super(sb, md, dt, mnt);
1074 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1076 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1079 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1080 LCONSOLE_WARN("Mounted %s\n", profilenm);
1084 } /* ll_fill_super */
1087 void lu_context_keys_dump(void);
1089 void ll_put_super(struct super_block *sb)
1091 struct config_llog_instance cfg, params_cfg;
1092 struct obd_device *obd;
1093 struct lustre_sb_info *lsi = s2lsi(sb);
1094 struct ll_sb_info *sbi = ll_s2sbi(sb);
1095 char *profilenm = get_profile_name(sb);
1096 int force = 1, next;
1099 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1101 ll_print_capa_stat(sbi);
1103 cfg.cfg_instance = sb;
1104 lustre_end_log(sb, profilenm, &cfg);
1106 params_cfg.cfg_instance = sb;
1107 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1109 if (sbi->ll_md_exp) {
1110 obd = class_exp2obd(sbi->ll_md_exp);
1112 force = obd->obd_force;
1115 /* We need to set force before the lov_disconnect in
1116 lustre_common_put_super, since l_d cleans up osc's as well. */
1119 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1121 obd->obd_force = force;
1126 /* Only if client_common_fill_super succeeded */
1127 client_common_put_super(sb);
1131 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1132 class_manual_cleanup(obd);
1135 if (sbi->ll_flags & LL_SBI_VERBOSE)
1136 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1139 class_del_profile(profilenm);
1141 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1142 bdi_destroy(&lsi->lsi_bdi);
1143 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1147 lsi->lsi_llsbi = NULL;
1149 lustre_common_put_super(sb);
1151 cfs_module_put(THIS_MODULE);
1154 } /* client_put_super */
1156 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1158 struct inode *inode = NULL;
1160 /* NOTE: we depend on atomic igrab() -bzzz */
1161 lock_res_and_lock(lock);
1162 if (lock->l_resource->lr_lvb_inode) {
1163 struct ll_inode_info * lli;
1164 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1165 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1166 inode = igrab(lock->l_resource->lr_lvb_inode);
1168 inode = lock->l_resource->lr_lvb_inode;
1169 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1170 D_WARNING, lock, "lr_lvb_inode %p is "
1171 "bogus: magic %08x",
1172 lock->l_resource->lr_lvb_inode,
1173 lli->lli_inode_magic);
1177 unlock_res_and_lock(lock);
1181 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1183 struct inode *inode = NULL;
1184 /* NOTE: we depend on atomic igrab() -bzzz */
1185 lock_res_and_lock(lock);
1186 if (lock->l_ast_data) {
1187 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1188 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1189 inode = igrab(lock->l_ast_data);
1191 inode = lock->l_ast_data;
1192 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1193 D_WARNING, lock, "l_ast_data %p is "
1194 "bogus: magic %08x", lock->l_ast_data,
1195 lli->lli_inode_magic);
1199 unlock_res_and_lock(lock);
1203 void ll_clear_inode(struct inode *inode)
1205 struct ll_inode_info *lli = ll_i2info(inode);
1206 struct ll_sb_info *sbi = ll_i2sbi(inode);
1209 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1210 inode->i_generation, inode);
1212 if (S_ISDIR(inode->i_mode)) {
1213 /* these should have been cleared in ll_file_release */
1214 LASSERT(lli->lli_opendir_key == NULL);
1215 LASSERT(lli->lli_sai == NULL);
1216 LASSERT(lli->lli_opendir_pid == 0);
1219 spin_lock(&lli->lli_lock);
1220 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1221 spin_unlock(&lli->lli_lock);
1222 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1224 LASSERT(!lli->lli_open_fd_write_count);
1225 LASSERT(!lli->lli_open_fd_read_count);
1226 LASSERT(!lli->lli_open_fd_exec_count);
1228 if (lli->lli_mds_write_och)
1229 ll_md_real_close(inode, FMODE_WRITE);
1230 if (lli->lli_mds_exec_och)
1231 ll_md_real_close(inode, FMODE_EXEC);
1232 if (lli->lli_mds_read_och)
1233 ll_md_real_close(inode, FMODE_READ);
1235 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1236 OBD_FREE(lli->lli_symlink_name,
1237 strlen(lli->lli_symlink_name) + 1);
1238 lli->lli_symlink_name = NULL;
1241 ll_xattr_cache_destroy(inode);
1243 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1244 LASSERT(lli->lli_posix_acl == NULL);
1245 if (lli->lli_remote_perms) {
1246 free_rmtperm_hash(lli->lli_remote_perms);
1247 lli->lli_remote_perms = NULL;
1250 #ifdef CONFIG_FS_POSIX_ACL
1251 else if (lli->lli_posix_acl) {
1252 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1253 LASSERT(lli->lli_remote_perms == NULL);
1254 posix_acl_release(lli->lli_posix_acl);
1255 lli->lli_posix_acl = NULL;
1258 lli->lli_inode_magic = LLI_INODE_DEAD;
1260 ll_clear_inode_capas(inode);
1261 if (!S_ISDIR(inode->i_mode))
1262 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1265 * XXX This has to be done before lsm is freed below, because
1266 * cl_object still uses inode lsm.
1268 cl_inode_fini(inode);
1269 lli->lli_has_smd = false;
1274 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1275 struct md_open_data **mod)
1277 struct lustre_md md;
1278 struct inode *inode = dentry->d_inode;
1279 struct ll_sb_info *sbi = ll_i2sbi(inode);
1280 struct ptlrpc_request *request = NULL;
1284 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1285 LUSTRE_OPC_ANY, NULL);
1286 if (IS_ERR(op_data))
1287 RETURN(PTR_ERR(op_data));
1289 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1292 ptlrpc_req_finished(request);
1293 if (rc == -ENOENT) {
1295 /* Unlinked special device node? Or just a race?
1296 * Pretend we done everything. */
1297 if (!S_ISREG(inode->i_mode) &&
1298 !S_ISDIR(inode->i_mode)) {
1299 ia_valid = op_data->op_attr.ia_valid;
1300 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1301 rc = simple_setattr(dentry, &op_data->op_attr);
1302 op_data->op_attr.ia_valid = ia_valid;
1304 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1305 CERROR("md_setattr fails: rc = %d\n", rc);
1310 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1311 sbi->ll_md_exp, &md);
1313 ptlrpc_req_finished(request);
1317 ia_valid = op_data->op_attr.ia_valid;
1318 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1319 * cache is not cleared yet. */
1320 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1321 rc = simple_setattr(dentry, &op_data->op_attr);
1322 op_data->op_attr.ia_valid = ia_valid;
1324 /* Extract epoch data if obtained. */
1325 op_data->op_handle = md.body->handle;
1326 op_data->op_ioepoch = md.body->ioepoch;
1328 ll_update_inode(inode, &md);
1329 ptlrpc_req_finished(request);
1334 /* Close IO epoch and send Size-on-MDS attribute update. */
1335 static int ll_setattr_done_writing(struct inode *inode,
1336 struct md_op_data *op_data,
1337 struct md_open_data *mod)
1339 struct ll_inode_info *lli = ll_i2info(inode);
1343 LASSERT(op_data != NULL);
1344 if (!S_ISREG(inode->i_mode))
1347 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1348 op_data->op_ioepoch, PFID(&lli->lli_fid));
1350 op_data->op_flags = MF_EPOCH_CLOSE;
1351 ll_done_writing_attr(inode, op_data);
1352 ll_pack_inode2opdata(inode, op_data, NULL);
1354 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1355 if (rc == -EAGAIN) {
1356 /* MDS has instructed us to obtain Size-on-MDS attribute
1357 * from OSTs and send setattr to back to MDS. */
1358 rc = ll_som_update(inode, op_data);
1360 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1366 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1368 struct obd_capa *capa;
1371 if (attr->ia_valid & ATTR_SIZE)
1372 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1374 capa = ll_mdscapa_get(inode);
1376 rc = cl_setattr_ost(inode, attr, capa);
1378 if (attr->ia_valid & ATTR_SIZE)
1379 ll_truncate_free_capa(capa);
1386 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1387 * object(s) determine the file size and mtime. Otherwise, the MDS will
1388 * keep these values until such a time that objects are allocated for it.
1389 * We do the MDS operations first, as it is checking permissions for us.
1390 * We don't to the MDS RPC if there is nothing that we want to store there,
1391 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1392 * going to do an RPC anyways.
1394 * If we are doing a truncate, we will send the mtime and ctime updates
1395 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1396 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1399 * In case of HSMimport, we only set attr on MDS.
1401 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1403 struct inode *inode = dentry->d_inode;
1404 struct ll_inode_info *lli = ll_i2info(inode);
1405 struct md_op_data *op_data = NULL;
1406 struct md_open_data *mod = NULL;
1407 bool file_is_released = false;
1408 int rc = 0, rc1 = 0;
1411 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1412 "valid %x, hsm_import %d\n",
1413 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1414 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1415 attr->ia_valid, hsm_import);
1417 if (attr->ia_valid & ATTR_SIZE) {
1418 /* Check new size against VFS/VM file size limit and rlimit */
1419 rc = inode_newsize_ok(inode, attr->ia_size);
1423 /* The maximum Lustre file size is variable, based on the
1424 * OST maximum object size and number of stripes. This
1425 * needs another check in addition to the VFS check above. */
1426 if (attr->ia_size > ll_file_maxbytes(inode)) {
1427 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1428 PFID(&lli->lli_fid), attr->ia_size,
1429 ll_file_maxbytes(inode));
1433 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1436 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1437 if (attr->ia_valid & TIMES_SET_FLAGS) {
1438 if (current_fsuid() != inode->i_uid &&
1439 !cfs_capable(CFS_CAP_FOWNER))
1443 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1444 if (attr->ia_valid & ATTR_CTIME) {
1445 attr->ia_ctime = CFS_CURRENT_TIME;
1446 attr->ia_valid |= ATTR_CTIME_SET;
1448 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1449 (attr->ia_valid & ATTR_ATIME)) {
1450 attr->ia_atime = CFS_CURRENT_TIME;
1451 attr->ia_valid |= ATTR_ATIME_SET;
1453 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1454 (attr->ia_valid & ATTR_MTIME)) {
1455 attr->ia_mtime = CFS_CURRENT_TIME;
1456 attr->ia_valid |= ATTR_MTIME_SET;
1459 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1460 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1461 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1462 cfs_time_current_sec());
1464 /* If we are changing file size, file content is modified, flag it. */
1465 if (attr->ia_valid & ATTR_SIZE) {
1466 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1467 spin_lock(&lli->lli_lock);
1468 lli->lli_flags |= LLIF_DATA_MODIFIED;
1469 spin_unlock(&lli->lli_lock);
1472 /* We always do an MDS RPC, even if we're only changing the size;
1473 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1475 OBD_ALLOC_PTR(op_data);
1476 if (op_data == NULL)
1479 if (!S_ISDIR(inode->i_mode)) {
1480 if (attr->ia_valid & ATTR_SIZE)
1481 inode_dio_write_done(inode);
1482 mutex_unlock(&inode->i_mutex);
1483 down_write(&lli->lli_trunc_sem);
1486 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1488 /* Open epoch for truncate. */
1489 if (exp_connect_som(ll_i2mdexp(inode)) &&
1490 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1491 op_data->op_flags = MF_EPOCH_OPEN;
1493 /* truncate on a released file must failed with -ENODATA,
1494 * so size must not be set on MDS for released file
1495 * but other attributes must be set
1497 if (S_ISREG(inode->i_mode)) {
1498 struct lov_stripe_md *lsm;
1501 ll_layout_refresh(inode, &gen);
1502 lsm = ccc_inode_lsm_get(inode);
1503 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1504 file_is_released = true;
1505 ccc_inode_lsm_put(inode, lsm);
1508 /* if not in HSM import mode, clear size attr for released file
1509 * we clear the attribute send to MDT in op_data, not the original
1510 * received from caller in attr which is used later to
1511 * decide return code */
1512 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1513 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1515 rc = ll_md_setattr(dentry, op_data, &mod);
1519 /* truncate failed (only when non HSM import), others succeed */
1520 if (file_is_released) {
1521 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1522 GOTO(out, rc = -ENODATA);
1527 /* RPC to MDT is sent, cancel data modification flag */
1528 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1529 spin_lock(&lli->lli_lock);
1530 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1531 spin_unlock(&lli->lli_lock);
1534 ll_ioepoch_open(lli, op_data->op_ioepoch);
1535 if (!S_ISREG(inode->i_mode))
1538 if (attr->ia_valid & (ATTR_SIZE |
1539 ATTR_ATIME | ATTR_ATIME_SET |
1540 ATTR_MTIME | ATTR_MTIME_SET)) {
1541 /* For truncate and utimes sending attributes to OSTs, setting
1542 * mtime/atime to the past will be performed under PW [0:EOF]
1543 * extent lock (new_size:EOF for truncate). It may seem
1544 * excessive to send mtime/atime updates to OSTs when not
1545 * setting times to past, but it is necessary due to possible
1546 * time de-synchronization between MDT inode and OST objects */
1547 rc = ll_setattr_ost(inode, attr);
1552 if (op_data->op_ioepoch) {
1553 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1557 ll_finish_md_op_data(op_data);
1559 if (!S_ISDIR(inode->i_mode)) {
1560 up_write(&lli->lli_trunc_sem);
1561 mutex_lock(&inode->i_mutex);
1562 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1563 inode_dio_wait(inode);
1566 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1567 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1572 int ll_setattr(struct dentry *de, struct iattr *attr)
1574 int mode = de->d_inode->i_mode;
1576 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1577 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1578 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1580 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1581 (ATTR_SIZE|ATTR_MODE)) &&
1582 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1583 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1584 !(attr->ia_mode & S_ISGID))))
1585 attr->ia_valid |= ATTR_FORCE;
1587 if ((mode & S_ISUID) &&
1588 !(attr->ia_mode & S_ISUID) &&
1589 !(attr->ia_valid & ATTR_KILL_SUID))
1590 attr->ia_valid |= ATTR_KILL_SUID;
1592 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1593 !(attr->ia_mode & S_ISGID) &&
1594 !(attr->ia_valid & ATTR_KILL_SGID))
1595 attr->ia_valid |= ATTR_KILL_SGID;
1597 return ll_setattr_raw(de, attr, false);
1600 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1601 __u64 max_age, __u32 flags)
1603 struct ll_sb_info *sbi = ll_s2sbi(sb);
1604 struct obd_statfs obd_osfs;
1608 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1610 CERROR("md_statfs fails: rc = %d\n", rc);
1614 osfs->os_type = sb->s_magic;
1616 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1617 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1619 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1620 flags |= OBD_STATFS_NODELAY;
1622 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1624 CERROR("obd_statfs fails: rc = %d\n", rc);
1628 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1629 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1632 osfs->os_bsize = obd_osfs.os_bsize;
1633 osfs->os_blocks = obd_osfs.os_blocks;
1634 osfs->os_bfree = obd_osfs.os_bfree;
1635 osfs->os_bavail = obd_osfs.os_bavail;
1637 /* If we don't have as many objects free on the OST as inodes
1638 * on the MDS, we reduce the total number of inodes to
1639 * compensate, so that the "inodes in use" number is correct.
1641 if (obd_osfs.os_ffree < osfs->os_ffree) {
1642 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1644 osfs->os_ffree = obd_osfs.os_ffree;
1649 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1651 struct super_block *sb = de->d_sb;
1652 struct obd_statfs osfs;
1653 __u64 fsid = huge_encode_dev(sb->s_dev);
1656 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1657 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1659 /* Some amount of caching on the client is allowed */
1660 rc = ll_statfs_internal(sb, &osfs,
1661 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1666 statfs_unpack(sfs, &osfs);
1668 /* We need to downshift for all 32-bit kernels, because we can't
1669 * tell if the kernel is being called via sys_statfs64() or not.
1670 * Stop before overflowing f_bsize - in which case it is better
1671 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1672 if (sizeof(long) < 8) {
1673 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1676 osfs.os_blocks >>= 1;
1677 osfs.os_bfree >>= 1;
1678 osfs.os_bavail >>= 1;
1682 sfs->f_blocks = osfs.os_blocks;
1683 sfs->f_bfree = osfs.os_bfree;
1684 sfs->f_bavail = osfs.os_bavail;
1685 sfs->f_fsid.val[0] = (__u32)fsid;
1686 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1690 void ll_inode_size_lock(struct inode *inode)
1692 struct ll_inode_info *lli;
1694 LASSERT(!S_ISDIR(inode->i_mode));
1696 lli = ll_i2info(inode);
1697 LASSERT(lli->lli_size_sem_owner != current);
1698 down(&lli->lli_size_sem);
1699 LASSERT(lli->lli_size_sem_owner == NULL);
1700 lli->lli_size_sem_owner = current;
1703 void ll_inode_size_unlock(struct inode *inode)
1705 struct ll_inode_info *lli;
1707 lli = ll_i2info(inode);
1708 LASSERT(lli->lli_size_sem_owner == current);
1709 lli->lli_size_sem_owner = NULL;
1710 up(&lli->lli_size_sem);
1713 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1715 struct ll_inode_info *lli = ll_i2info(inode);
1716 struct mdt_body *body = md->body;
1717 struct lov_stripe_md *lsm = md->lsm;
1718 struct ll_sb_info *sbi = ll_i2sbi(inode);
1720 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1722 if (!lli->lli_has_smd &&
1723 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1724 cl_file_inode_init(inode, md);
1726 lli->lli_maxbytes = lsm->lsm_maxbytes;
1727 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1728 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1731 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1732 if (body->valid & OBD_MD_FLRMTPERM)
1733 ll_update_remote_perm(inode, md->remote_perm);
1735 #ifdef CONFIG_FS_POSIX_ACL
1736 else if (body->valid & OBD_MD_FLACL) {
1737 spin_lock(&lli->lli_lock);
1738 if (lli->lli_posix_acl)
1739 posix_acl_release(lli->lli_posix_acl);
1740 lli->lli_posix_acl = md->posix_acl;
1741 spin_unlock(&lli->lli_lock);
1744 inode->i_ino = cl_fid_build_ino(&body->fid1,
1745 sbi->ll_flags & LL_SBI_32BIT_API);
1746 inode->i_generation = cl_fid_build_gen(&body->fid1);
1748 if (body->valid & OBD_MD_FLATIME) {
1749 if (body->atime > LTIME_S(inode->i_atime))
1750 LTIME_S(inode->i_atime) = body->atime;
1751 lli->lli_lvb.lvb_atime = body->atime;
1753 if (body->valid & OBD_MD_FLMTIME) {
1754 if (body->mtime > LTIME_S(inode->i_mtime)) {
1755 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1756 "to "LPU64"\n", inode->i_ino,
1757 LTIME_S(inode->i_mtime), body->mtime);
1758 LTIME_S(inode->i_mtime) = body->mtime;
1760 lli->lli_lvb.lvb_mtime = body->mtime;
1762 if (body->valid & OBD_MD_FLCTIME) {
1763 if (body->ctime > LTIME_S(inode->i_ctime))
1764 LTIME_S(inode->i_ctime) = body->ctime;
1765 lli->lli_lvb.lvb_ctime = body->ctime;
1767 if (body->valid & OBD_MD_FLMODE)
1768 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1769 if (body->valid & OBD_MD_FLTYPE)
1770 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1771 LASSERT(inode->i_mode != 0);
1772 if (S_ISREG(inode->i_mode)) {
1773 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1775 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1777 if (body->valid & OBD_MD_FLUID)
1778 inode->i_uid = body->uid;
1779 if (body->valid & OBD_MD_FLGID)
1780 inode->i_gid = body->gid;
1781 if (body->valid & OBD_MD_FLFLAGS)
1782 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1783 if (body->valid & OBD_MD_FLNLINK)
1784 set_nlink(inode, body->nlink);
1785 if (body->valid & OBD_MD_FLRDEV)
1786 inode->i_rdev = old_decode_dev(body->rdev);
1788 if (body->valid & OBD_MD_FLID) {
1789 /* FID shouldn't be changed! */
1790 if (fid_is_sane(&lli->lli_fid)) {
1791 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1792 "Trying to change FID "DFID
1793 " to the "DFID", inode %lu/%u(%p)\n",
1794 PFID(&lli->lli_fid), PFID(&body->fid1),
1795 inode->i_ino, inode->i_generation, inode);
1797 lli->lli_fid = body->fid1;
1800 LASSERT(fid_seq(&lli->lli_fid) != 0);
1802 if (body->valid & OBD_MD_FLSIZE) {
1803 if (exp_connect_som(ll_i2mdexp(inode)) &&
1804 S_ISREG(inode->i_mode)) {
1805 struct lustre_handle lockh;
1808 /* As it is possible a blocking ast has been processed
1809 * by this time, we need to check there is an UPDATE
1810 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1812 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1813 &lockh, LDLM_FL_CBPENDING,
1817 if (lli->lli_flags & (LLIF_DONE_WRITING |
1818 LLIF_EPOCH_PENDING |
1820 CERROR("ino %lu flags %u still has "
1821 "size authority! do not trust "
1822 "the size got from MDS\n",
1823 inode->i_ino, lli->lli_flags);
1825 /* Use old size assignment to avoid
1826 * deadlock bz14138 & bz14326 */
1827 i_size_write(inode, body->size);
1828 spin_lock(&lli->lli_lock);
1829 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1830 spin_unlock(&lli->lli_lock);
1832 ldlm_lock_decref(&lockh, mode);
1835 /* Use old size assignment to avoid
1836 * deadlock bz14138 & bz14326 */
1837 i_size_write(inode, body->size);
1839 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1840 inode->i_ino, (unsigned long long)body->size);
1843 if (body->valid & OBD_MD_FLBLOCKS)
1844 inode->i_blocks = body->blocks;
1847 if (body->valid & OBD_MD_FLMDSCAPA) {
1848 LASSERT(md->mds_capa);
1849 ll_add_capa(inode, md->mds_capa);
1851 if (body->valid & OBD_MD_FLOSSCAPA) {
1852 LASSERT(md->oss_capa);
1853 ll_add_capa(inode, md->oss_capa);
1856 if (body->valid & OBD_MD_TSTATE) {
1857 if (body->t_state & MS_RESTORE)
1858 lli->lli_flags |= LLIF_FILE_RESTORING;
1862 void ll_read_inode2(struct inode *inode, void *opaque)
1864 struct lustre_md *md = opaque;
1865 struct ll_inode_info *lli = ll_i2info(inode);
1868 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1869 PFID(&lli->lli_fid), inode);
1871 LASSERT(!lli->lli_has_smd);
1873 /* Core attributes from the MDS first. This is a new inode, and
1874 * the VFS doesn't zero times in the core inode so we have to do
1875 * it ourselves. They will be overwritten by either MDS or OST
1876 * attributes - we just need to make sure they aren't newer. */
1877 LTIME_S(inode->i_mtime) = 0;
1878 LTIME_S(inode->i_atime) = 0;
1879 LTIME_S(inode->i_ctime) = 0;
1881 ll_update_inode(inode, md);
1883 /* OIDEBUG(inode); */
1885 /* initializing backing dev info. */
1886 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1889 if (S_ISREG(inode->i_mode)) {
1890 struct ll_sb_info *sbi = ll_i2sbi(inode);
1891 inode->i_op = &ll_file_inode_operations;
1892 inode->i_fop = sbi->ll_fop;
1893 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1895 } else if (S_ISDIR(inode->i_mode)) {
1896 inode->i_op = &ll_dir_inode_operations;
1897 inode->i_fop = &ll_dir_operations;
1899 } else if (S_ISLNK(inode->i_mode)) {
1900 inode->i_op = &ll_fast_symlink_inode_operations;
1903 inode->i_op = &ll_special_inode_operations;
1905 init_special_inode(inode, inode->i_mode,
1912 void ll_delete_inode(struct inode *inode)
1914 struct cl_inode_info *lli = cl_i2info(inode);
1917 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1918 /* discard all dirty pages before truncating them, required by
1919 * osc_extent implementation at LU-1030. */
1920 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1921 CL_FSYNC_DISCARD, 1);
1923 truncate_inode_pages(&inode->i_data, 0);
1925 /* Workaround for LU-118 */
1926 if (inode->i_data.nrpages) {
1927 TREE_READ_LOCK_IRQ(&inode->i_data);
1928 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1929 LASSERTF(inode->i_data.nrpages == 0,
1930 "inode=%lu/%u(%p) nrpages=%lu, see "
1931 "http://jira.whamcloud.com/browse/LU-118\n",
1932 inode->i_ino, inode->i_generation, inode,
1933 inode->i_data.nrpages);
1935 /* Workaround end */
1937 #ifdef HAVE_SBOPS_EVICT_INODE
1938 ll_clear_inode(inode);
1945 int ll_iocontrol(struct inode *inode, struct file *file,
1946 unsigned int cmd, unsigned long arg)
1948 struct ll_sb_info *sbi = ll_i2sbi(inode);
1949 struct ptlrpc_request *req = NULL;
1954 case FSFILT_IOC_GETFLAGS: {
1955 struct mdt_body *body;
1956 struct md_op_data *op_data;
1958 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1959 0, 0, LUSTRE_OPC_ANY,
1961 if (IS_ERR(op_data))
1962 RETURN(PTR_ERR(op_data));
1964 op_data->op_valid = OBD_MD_FLFLAGS;
1965 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1966 ll_finish_md_op_data(op_data);
1968 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1972 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1974 flags = body->flags;
1976 ptlrpc_req_finished(req);
1978 RETURN(put_user(flags, (int *)arg));
1980 case FSFILT_IOC_SETFLAGS: {
1981 struct lov_stripe_md *lsm;
1982 struct obd_info oinfo = { { { 0 } } };
1983 struct md_op_data *op_data;
1985 if (get_user(flags, (int *)arg))
1988 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1989 LUSTRE_OPC_ANY, NULL);
1990 if (IS_ERR(op_data))
1991 RETURN(PTR_ERR(op_data));
1993 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1994 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1995 rc = md_setattr(sbi->ll_md_exp, op_data,
1996 NULL, 0, NULL, 0, &req, NULL);
1997 ll_finish_md_op_data(op_data);
1998 ptlrpc_req_finished(req);
2002 inode->i_flags = ll_ext_to_inode_flags(flags);
2004 lsm = ccc_inode_lsm_get(inode);
2005 if (!lsm_has_objects(lsm)) {
2006 ccc_inode_lsm_put(inode, lsm);
2010 OBDO_ALLOC(oinfo.oi_oa);
2012 ccc_inode_lsm_put(inode, lsm);
2016 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2017 oinfo.oi_oa->o_flags = flags;
2018 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2020 oinfo.oi_capa = ll_mdscapa_get(inode);
2021 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2022 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2023 capa_put(oinfo.oi_capa);
2024 OBDO_FREE(oinfo.oi_oa);
2025 ccc_inode_lsm_put(inode, lsm);
2027 if (rc && rc != -EPERM && rc != -EACCES)
2028 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2039 int ll_flush_ctx(struct inode *inode)
2041 struct ll_sb_info *sbi = ll_i2sbi(inode);
2043 CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2045 obd_set_info_async(NULL, sbi->ll_md_exp,
2046 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2048 obd_set_info_async(NULL, sbi->ll_dt_exp,
2049 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2054 /* umount -f client means force down, don't save state */
2055 void ll_umount_begin(struct super_block *sb)
2057 struct ll_sb_info *sbi = ll_s2sbi(sb);
2058 struct obd_device *obd;
2059 struct obd_ioctl_data *ioc_data;
2062 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2063 sb->s_count, atomic_read(&sb->s_active));
2065 obd = class_exp2obd(sbi->ll_md_exp);
2067 CERROR("Invalid MDC connection handle "LPX64"\n",
2068 sbi->ll_md_exp->exp_handle.h_cookie);
2074 obd = class_exp2obd(sbi->ll_dt_exp);
2076 CERROR("Invalid LOV connection handle "LPX64"\n",
2077 sbi->ll_dt_exp->exp_handle.h_cookie);
2083 OBD_ALLOC_PTR(ioc_data);
2085 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2086 sizeof *ioc_data, ioc_data, NULL);
2088 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2089 sizeof *ioc_data, ioc_data, NULL);
2091 OBD_FREE_PTR(ioc_data);
2094 /* Really, we'd like to wait until there are no requests outstanding,
2095 * and then continue. For now, we just invalidate the requests,
2096 * schedule() and sleep one second if needed, and hope.
2102 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2104 struct ll_sb_info *sbi = ll_s2sbi(sb);
2105 char *profilenm = get_profile_name(sb);
2109 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2110 read_only = *flags & MS_RDONLY;
2111 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2112 sizeof(KEY_READ_ONLY),
2113 KEY_READ_ONLY, sizeof(read_only),
2116 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2117 profilenm, read_only ?
2118 "read-only" : "read-write", err);
2123 sb->s_flags |= MS_RDONLY;
2125 sb->s_flags &= ~MS_RDONLY;
2127 if (sbi->ll_flags & LL_SBI_VERBOSE)
2128 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2129 read_only ? "read-only" : "read-write");
2134 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2135 struct super_block *sb, struct lookup_intent *it)
2137 struct ll_sb_info *sbi = NULL;
2138 struct lustre_md md;
2142 LASSERT(*inode || sb);
2143 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2144 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2145 sbi->ll_md_exp, &md);
2150 ll_update_inode(*inode, &md);
2152 LASSERT(sb != NULL);
2155 * At this point server returns to client's same fid as client
2156 * generated for creating. So using ->fid1 is okay here.
2158 LASSERT(fid_is_sane(&md.body->fid1));
2160 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2161 sbi->ll_flags & LL_SBI_32BIT_API),
2163 if (*inode == NULL || IS_ERR(*inode)) {
2164 #ifdef CONFIG_FS_POSIX_ACL
2166 posix_acl_release(md.posix_acl);
2167 md.posix_acl = NULL;
2170 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2172 CERROR("new_inode -fatal: rc %d\n", rc);
2177 /* Handling piggyback layout lock.
2178 * Layout lock can be piggybacked by getattr and open request.
2179 * The lsm can be applied to inode only if it comes with a layout lock
2180 * otherwise correct layout may be overwritten, for example:
2181 * 1. proc1: mdt returns a lsm but not granting layout
2182 * 2. layout was changed by another client
2183 * 3. proc2: refresh layout and layout lock granted
2184 * 4. proc1: to apply a stale layout */
2185 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2186 struct lustre_handle lockh;
2187 struct ldlm_lock *lock;
2189 lockh.cookie = it->d.lustre.it_lock_handle;
2190 lock = ldlm_handle2lock(&lockh);
2191 LASSERT(lock != NULL);
2192 if (ldlm_has_layout(lock)) {
2193 struct cl_object_conf conf;
2195 memset(&conf, 0, sizeof(conf));
2196 conf.coc_opc = OBJECT_CONF_SET;
2197 conf.coc_inode = *inode;
2198 conf.coc_lock = lock;
2199 conf.u.coc_md = &md;
2200 (void)ll_layout_conf(*inode, &conf);
2202 LDLM_LOCK_PUT(lock);
2207 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2208 md_free_lustre_md(sbi->ll_md_exp, &md);
2212 int ll_obd_statfs(struct inode *inode, void *arg)
2214 struct ll_sb_info *sbi = NULL;
2215 struct obd_export *exp;
2217 struct obd_ioctl_data *data = NULL;
2222 if (!inode || !(sbi = ll_i2sbi(inode)))
2223 GOTO(out_statfs, rc = -EINVAL);
2225 rc = obd_ioctl_getdata(&buf, &len, arg);
2227 GOTO(out_statfs, rc);
2230 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2231 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2232 GOTO(out_statfs, rc = -EINVAL);
2234 if (data->ioc_inllen1 != sizeof(__u32) ||
2235 data->ioc_inllen2 != sizeof(__u32) ||
2236 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2237 data->ioc_plen2 != sizeof(struct obd_uuid))
2238 GOTO(out_statfs, rc = -EINVAL);
2240 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2241 if (type & LL_STATFS_LMV)
2242 exp = sbi->ll_md_exp;
2243 else if (type & LL_STATFS_LOV)
2244 exp = sbi->ll_dt_exp;
2246 GOTO(out_statfs, rc = -ENODEV);
2248 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2249 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2251 GOTO(out_statfs, rc);
2254 obd_ioctl_freedata(buf, len);
2258 int ll_process_config(struct lustre_cfg *lcfg)
2262 struct lprocfs_static_vars lvars;
2266 lprocfs_llite_init_vars(&lvars);
2268 /* The instance name contains the sb: lustre-client-aacfe000 */
2269 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2270 if (!ptr || !*(++ptr))
2272 if (sscanf(ptr, "%lx", &x) != 1)
2275 /* This better be a real Lustre superblock! */
2276 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2278 /* Note we have not called client_common_fill_super yet, so
2279 proc fns must be able to handle that! */
2280 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2287 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2288 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2289 struct inode *i1, struct inode *i2,
2290 const char *name, int namelen,
2291 int mode, __u32 opc, void *data)
2293 LASSERT(i1 != NULL);
2295 if (namelen > ll_i2sbi(i1)->ll_namelen)
2296 return ERR_PTR(-ENAMETOOLONG);
2298 if (op_data == NULL)
2299 OBD_ALLOC_PTR(op_data);
2301 if (op_data == NULL)
2302 return ERR_PTR(-ENOMEM);
2304 ll_i2gids(op_data->op_suppgids, i1, i2);
2305 op_data->op_fid1 = *ll_inode2fid(i1);
2306 op_data->op_capa1 = ll_mdscapa_get(i1);
2309 op_data->op_fid2 = *ll_inode2fid(i2);
2310 op_data->op_capa2 = ll_mdscapa_get(i2);
2312 fid_zero(&op_data->op_fid2);
2313 op_data->op_capa2 = NULL;
2316 op_data->op_name = name;
2317 op_data->op_namelen = namelen;
2318 op_data->op_mode = mode;
2319 op_data->op_mod_time = cfs_time_current_sec();
2320 op_data->op_fsuid = current_fsuid();
2321 op_data->op_fsgid = current_fsgid();
2322 op_data->op_cap = cfs_curproc_cap_pack();
2323 op_data->op_bias = 0;
2324 op_data->op_cli_flags = 0;
2325 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2326 filename_is_volatile(name, namelen, NULL))
2327 op_data->op_bias |= MDS_CREATE_VOLATILE;
2328 op_data->op_opc = opc;
2329 op_data->op_mds = 0;
2330 op_data->op_data = data;
2332 /* If the file is being opened after mknod() (normally due to NFS)
2333 * try to use the default stripe data from parent directory for
2334 * allocating OST objects. Try to pass the parent FID to MDS. */
2335 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2336 !ll_i2info(i2)->lli_has_smd) {
2337 struct ll_inode_info *lli = ll_i2info(i2);
2339 spin_lock(&lli->lli_lock);
2340 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2341 op_data->op_fid1 = lli->lli_pfid;
2342 spin_unlock(&lli->lli_lock);
2343 /** We ignore parent's capability temporary. */
2346 /* When called by ll_setattr_raw, file is i1. */
2347 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2348 op_data->op_bias |= MDS_DATA_MODIFIED;
2353 void ll_finish_md_op_data(struct md_op_data *op_data)
2355 capa_put(op_data->op_capa1);
2356 capa_put(op_data->op_capa2);
2357 OBD_FREE_PTR(op_data);
2360 #ifdef HAVE_SUPEROPS_USE_DENTRY
2361 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2363 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2366 struct ll_sb_info *sbi;
2368 #ifdef HAVE_SUPEROPS_USE_DENTRY
2369 LASSERT((seq != NULL) && (dentry != NULL));
2370 sbi = ll_s2sbi(dentry->d_sb);
2372 LASSERT((seq != NULL) && (vfs != NULL));
2373 sbi = ll_s2sbi(vfs->mnt_sb);
2376 if (sbi->ll_flags & LL_SBI_NOLCK)
2377 seq_puts(seq, ",nolock");
2379 if (sbi->ll_flags & LL_SBI_FLOCK)
2380 seq_puts(seq, ",flock");
2382 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2383 seq_puts(seq, ",localflock");
2385 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2386 seq_puts(seq, ",user_xattr");
2388 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2389 seq_puts(seq, ",lazystatfs");
2391 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2392 seq_puts(seq, ",user_fid2path");
2398 * Get obd name by cmd, and copy out to user space
2400 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2402 struct ll_sb_info *sbi = ll_i2sbi(inode);
2403 struct obd_device *obd;
2406 if (cmd == OBD_IOC_GETDTNAME)
2407 obd = class_exp2obd(sbi->ll_dt_exp);
2408 else if (cmd == OBD_IOC_GETMDNAME)
2409 obd = class_exp2obd(sbi->ll_md_exp);
2416 if (copy_to_user((void *)arg, obd->obd_name,
2417 strlen(obd->obd_name) + 1))
2424 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2425 * fsname will be returned in this buffer; otherwise, a static buffer will be
2426 * used to store the fsname and returned to caller.
2428 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2430 static char fsname_static[MTI_NAME_MAXLEN];
2431 struct lustre_sb_info *lsi = s2lsi(sb);
2436 /* this means the caller wants to use static buffer
2437 * and it doesn't care about race. Usually this is
2438 * in error reporting path */
2439 buf = fsname_static;
2440 buflen = sizeof(fsname_static);
2443 len = strlen(lsi->lsi_lmd->lmd_profile);
2444 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2445 if (ptr && (strcmp(ptr, "-client") == 0))
2448 if (unlikely(len >= buflen))
2450 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2456 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2463 p.mnt = current->fs->root.mnt;
2465 path = d_path(&p, buf, bufsize);
2470 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2472 char *buf, *path = NULL;
2473 struct dentry *dentry = NULL;
2474 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2476 /* this can be called inside spin lock so use GFP_ATOMIC. */
2477 buf = (char *)__get_free_page(GFP_ATOMIC);
2479 dentry = d_find_alias(page->mapping->host);
2481 path = ll_d_path(dentry, buf, PAGE_SIZE);
2484 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2485 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2486 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2487 PFID(&obj->cob_header.coh_lu.loh_fid),
2488 (path && !IS_ERR(path)) ? path : "", ioret);
2494 free_page((unsigned long)buf);